OpenAI's GPT-5.5 Advances Enterprise Workflows and Enters Cyber Defense

Databricks is integrating GPT-5.5 into its customer agent workflows after the model achieved state-of-the-art results on the OfficeQA Pro benchmark. [2] The OfficeQA Pro benchmark tests how models manage grounded reasoning, parsing, and retrieval tasks involving long-context documents, legacy files, and scanned PDFs. [2] In an agent-harness setting, GPT-5.5 became the first model to exceed 50 percent accuracy on this evaluation. [2] Additionally, the model demonstrated a 46 percent reduction in errors when compared directly to the previous GPT-5.4 model. [2] Research engineer Arnav Singhvi noted that Codex powered by the 5.5 model is currently state-of-the-art among available models and agents. [2]

The primary gains for Databricks occurred in parsing-heavy workflows, as GPT-5.5 showed improved ability to correctly parse digits from scanned PDFs and older documents compared to GPT-5.4. [2] The updated model also demonstrated better orchestration during multi-step tasks by avoiding unnecessary search detours that previously caused inefficient trajectories. [2] Databricks currently offers GPT-5.5 through the AI Unity Gateway, allowing customers to utilize the model within workflows constructed using the Agent Supervisor API and AgentBricks. [2] Alongside these enterprise deployments, OpenAI is also pushing GPT-5.5 into cyber defense applications through a new effort called the "Daybreak" initiative. [1]